Zenith Grant Awardee

Felix Binder

IQOQI Vienna, Austrian Academy of Sciences

Co-Investigators

Kavan Modi, Monash University; Felix Pollock, Monash University

Project Title

Predictive Quantum Intelligence under Physical Constraints

Project Summary

If we want to understand something about the world we have to carefully probe and observe it. For a physical phenomenon of interest – the weather, say – we can then form a model and use it to make predictions: when the sky is overcast, there is a higher chance of snowfall than on a sunny day. Arriving at such predictions requires memory and reasoning. We will only conclude that a dark sky indicates a chance of snowfall if we have previously experienced the two to be correlated. We understand intelligence as a combination of memory and the ability to think. But how much intelligence is required to make a good prediction? And what do we do if our intelligence is too limited to comprehend the full picture? In this research, we study how much an intelligent being can learn about the quantum world and what quantum aspects will remain hidden from us if we are limited in the ways in which we can interact with it – for instance, because our human senses are too limited to discern quantum phenomena with the naked eye.

Technical Abstract

Intelligence is the capacity to make sense of input from the external world. An intelligent being or agent is able not only to perceive their exterior environment but to interact with it in a structured way and to predict its future behaviour. In this research project, we investigate how much intelligence – here: the combination of memory and processing capacity – an agent requires to establish a model of a non-Markovian quantum stochastic process from observation. Putting the agent's capacity to probe the environment at the center of the description we use a quantum comb to describe both environment and agent. This allows us to derive lower bounds on the intelligence required of the agent in order to fully comprehend the process by means of a predictive model. In a second step, we consider agents who are constrained in terms of memory capacity or access to physical degrees of freedom and ask which features of a process, such as quantumness, they may still discern.